Download Prediction and Measurement of Temperature Fields in Silicon

Survey
yes no Was this document useful for you?
   Thank you for your participation!

* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project

Document related concepts

Giant magnetoresistance wikipedia , lookup

UniPro protocol stack wikipedia , lookup

Integrated circuit wikipedia , lookup

Nanofluidic circuitry wikipedia , lookup

Power MOSFET wikipedia , lookup

Thermal copper pillar bump wikipedia , lookup

Thermal runaway wikipedia , lookup

Lumped element model wikipedia , lookup

Transcript
K. E. Goodson ~
Assoc, Mere. ASME
M. I. Flik 2
Department of MechanicalEngineering.
L.T. Su 3
D. A. Antoniadis
Department of Electrical Engineeringand
Computer Science.
Massachusetts Institute of Technology,
Cambridge, MA 02139
1
Prediction and Measurement of
Temperature Fields in Siliconon-Insulator Electronic Circuits
Field-effect transistors (FETs) in conventional electronic circuits are in contact with
the high-thermal-conductivity substrate. In contrast, FETs in novel silicon-on-insulator
(SOl) circuits are separated from the substrate by a thermally resistive silicon-dioxide
layer. The layer improves the electrical performance of SO1 circuits. But it impedes
conduction cooling of transistors and interconnects, degrading circuit reliability. This
work develops a technique for measuring the channel temperature of SOI FETs. Data
agree well with the predictions of an analytical thermal model. The channel and interconnect temperatures depend strongly on the device and silicon-dioxide layer thicknesses and the channel-interconnect separation. This research facilitates the thermal
design of SO1 FETs to improve circuit figures of merit, e.g., the median time to failure
(MTF) of FET-interconnect contacts.
Introduction
The performance and reliability of electronic circuits are affected by temperature fields in transistors and interconnects. As
transistor dimensions decrease, thermal conduction within a few
micrometers of these heat sources governs an increasing fraction
of the transistor-to-coolant temperature difference in an electronic system. This is very important in novel silicon-on-insulator
(SOI) electronic circuits, where transistors are separated from
the substrate by a thermally resistive silicon-dioxide layer, often
fabricated by implanting oxygen ions into a single-crystal silicon
wafer. Figure 1 is a cross section of a SOI field-effect transistor
(FET), whose common dimensions are given in Table 1. Almost
all of the device power is dissipated in the channel. The electrically insulating implanted layer prevents latchup between devices and reduces the parasitic capacitance of the transistor due
to the substrate, facilitating faster circuits (Colinge, 1991). But
the implanted silicon-dioxide layer has a low thermal conductivity, and impedes conduction cooling of the channel through the
substrate. The SOI circuit designer must know the resulting temperature rise in interconnects and devices.
McDaid et al. (1989) showed that the temperature rise decreased the drain current of a SO1 FET for given gate and drain
voltages by assuming one-dimensional heat conduction through
the implanted layer from an isothermal FET. Goodson and Flik
(1992) predicted the temperature field in a SOI FET by treating
the source, drain, gate, and interconnects as cooling fins for the
Joule-heated channel. They showed that the temperature varies
significantly within the FET, and indicated that the temperature
rise could reduce the electromigration-limitedreliability of interconnects. For application of SO1 circuits below 77 K, e.g., in
hybrid superconductor-semiconductor circuits, phonon-boundary
scattering was shown to influence the temperature field strongly
in FETs.
Experimental confirmation of the analysis of Goodson and Flik
(1992) requires a technique for measuring temperature locally in
transistors with spatial resolution comparable to the channel
length, 2Lg, which can be smaller than 0.5 #m. Lifka and Woer-
Stanlbrd University, Mechanical Engineering Department, Stan|brd, CA 94305.
2 Behr GmbH & Co., Mauserstrasse 3, 70469 Stuttgart, Federal Republic of' Germany.
IBM Corporation, East Fishkill Facility, 1580 Route 52 (E40), Hopewell Junction, NY 12533.
Contributed by the Heat Transfer Division and presented at the ASME/AIChE
National Heat Transfer Conference, Atlanta, Georgia, August 8-11, 1993. Manuscript received by the Heat Transfer Division September 1993; revision received
May 1994. Keywords: Conduction, Electronic Equipment, Measurement Techniques. Associate Technical Editor: R. Viskanta.
574 / Vol. 117, AUGUST 1995
lee (1990) estimated the transistor-to-substrate thermal resistance for SOl circuits through the local melting of a coating, an
approach that lacks the needed spatial resolution. Bunyan et al.
(1992) used noise thermometry to measure temperatures in SOl
transistors. This approach requires experimental structures very
different from those in a circuit, yielding an impact on the temperature field that needs to be assessed. For non-SO1 semiconductor devices, Brugger (1991) and Ostermeier et al. (1992)
performed temperature measurements with submicrometer spatial resolution using micro-Raman spectroscopy. But the impact
of the incident radiation on the performance of transistors is not
known. The promising method of Majumdar et al. (1993) and
Lai et al. (1993) uses an atomic force microscope to determine
local temperature fields in circuits. Future research will almost
certainly make this method effective for transistors with channel
lengths much less than 1 #m.
This work develops a technique to measure the channel temperature of SO1 FETs with a resolution in the direction of current
flow of 0.32 #m, the FET channel length. The gate serves as an
electrical-resistance thermometer for the channel temperature.
This approach was used by Mautry and Trager (1990) for bulk
(non-SOI) circuits, in which the characteristic length scale of the
temperature field is much larger than FET dimensions. The gate
is not isothermal in SOl circuits, in which the high thermal resistance of the implanted layer causes a large fraction of the channel-to-coolant temperature difference to occur within transistors.
In the present work, thermal analysis yields the average channel
temperature from the electrical resistance measured along the
nonisothermal gate. Room-temperature data are compared with
predictions of the analysis of Goodson and Flik (1992). Analysis
estimates the impact of the channel-interconnect separation and
the implanted-silicon-dioxide layer thicknesses on the reliability
of FET-interconnect contacts.
This work helps to determine the effect of the implanted-layer
thermal resistance on the practical potential of SOl technology.
The experimental method developed here investigates thermal
conduction processes within and very near transistors and interconnects. These conduction processes must be understood before
transistor design based on thermal analysis, i.e., device-level
thermal design, can accompany traditional electrical design of
devices to yield circuits of optimal performance and reliability.
2
Thermal Analysis
SOI circuits are made of materials with very different thermal
conductivities, shown in Table 2. Goodson and Flik (1992) used
the difference in the conductivities to develop a simple thermal
Copyright © 1995 by ASME
Transactions of the ASME
Downloaded 21 Nov 2008 to 171.64.49.29. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
model for thermal conduction in SO1 circuits. Section 2.1 reviews
the analysis and Section 2.2 applies it to the experimental structure.
2,1 Steady-State SO1 F E T T h e r m a l Model. The model of
Goodson and Flik (1992) is for steady-state FET operation,
which is the case in most measurements of FET electrical properties. The time required for steady state to be achieved is near
(d,,)-~/(k,,/C,,), where k,,/C,, is the thermal diffusivity of silicon
dioxide, yielding about 200 ns at 300 K. The model is also useful
for the case of steady-periodic power dissipation in clock-driven
circuits at points separated from the channel heater by at least
one thermal penetration depth. The thermal penetration depth in
the silicon source and drain is approximately (rkd/C,t)~/2 = 0.4
#m, where k J C , i is the thermal diffusivity of heavily doped silicon, and r = 5 ns is a typical clock period.
Heat flow from the tops of devices and interconnects is shown
in Section 2.2 to be negligible. The channel was modeled as an
isothermal heating source. This neglects the complex distribution
of heating intensity in the channel, yielding a small error in the
average channel temperature that is estimated at the end of Section 2.1. Variations in the temperature o f the substrate-silicon
dioxide interface are small compared to the channel-temperature
rise. This interface was assumed to have the uniform temperature T0.
The source, drain, gate, and interconnects were modeled by
Goodson and Flik (1992) as one-dimensional cooling fins for the
Z
2Lg
Ld
GATE
Fig, 1
Cross section of a silicon-on-insulator (SOl) field-effect transistor
(FET)
channel, arranged as shown in Fig. 2. The heat loss per unit fin
length was assumed to be hrwr( ~ - To), where T t is the local
fin temperature and wl is the fin width'in the direction normal to
heat flow. The heat transfer coefficient is ht = ~k,,/d~,,, where d~,,
is the thickness and k,,/d~,, is the inverse volume resistance of the
silicon-dioxide layer beneath. The dimensionless function ~P was
derived by Goodson and Flik (1992) to account for two-dimensional conduction in the silicon dioxide. The function depends
Nomenclature
A = area, m 2
k,. = channel thermal conductivity,
A t = fin cross-sectional area, m 2
a = characteristic length of test
structure, m
C = specific heat at constant volume per unit volume,
J m 3 K-i
c/, = specific heat at constant pressure per unit mass, J kg-I K ~
d = layer thickness, m
d,. = total thickness of thermally
grown and CVD silicon dioxide, m
de,, = thickness of silicon-dioxide
layer between gate and channel, m
d,, = thickness of implanted-silicondioxide layer, m
dl,, = thickness of silicon dioxide between fin and substrate, m
E,, = electromigration activation energy, J
Fc,(Y) = shape function for temperature
distribution in gate
G = channel-to-air thermal conductance, W K-~
g = acceleration due to gravity,
m s -2
h = heat transfer coefficient, W
m =K--~
1,7 = drain current, A
Ic = gate current, A
J = current density, A m -2
K,, = electromigration constant, s
k = thermal conductivity,
W m tK-t
ke = Boltzmann constant =
1.38 × 1 0 -23 J K i
k,, =
k~ =
L,i =
L~ =
L,,, =
MTF =
m =
P =
P,v~ =
R, =
Re; =
Ru(T) =
T =
T,. =
T+;(Y) =
ATu =
T,......... =
T0 =
Vc; =
VGS =
Vos =
l) =
Journal of Heat Transfer
Wm+' K ~
thermal conductivity of silicon
dioxide, W m-~ K ~
thermal conductivity of lightly
doped silicon, W m-~ K '
separation between gate and
metal interconnect, m
half-length o f gate in X direction, m
half-length o f interconnect between devices in X direction,
m
median time to failure, s
(h/kd) ~12 = inverse thermal
healing length o f fin, m '
device power, W
time-averaged device power,
W
channel-to-substrate thermal
resistance = 1/G, K W -t
electrical resistance of gate in
Y d!rection, f~
gate-electrical-resistance calibration function, 9t
temperature, K
average channel temperature,
K
gate temperature distribution,
K
average temperature rise in
gate, K
maximum interconnect temperature, K
substrate temperature, K
voltage drop along gate, V
gate-source voltage drop, V
drain-source voltage drop, V
air velocity, m s E
w t -- fin width in X - Y plane,
m
w,~ = channel width in Y direction, m
w,. = separation between
channel and gate contacts in Y direction, m
w,,, = width o f metal interconnect in Y direction, m
X = coordinate in plane of
substrate, m
Y = coordinate in plane of
substrate, m
Z = coordinate normal to
plane of substrate, m
Z~, Z2, Z3, Z4 = constants, Eqs. (1)-(4)
/3 = coefficient of thermal
expansion, K
e = emissivity
/z = viscosity, kg m - t s '
p = mass density, kg m 3
~r = S t e f a n - B o l t z m a n n constant = 5.67 × 10 ~
W m 2K 4
r = clock period, s
• = two-dimensional conduction function
Subscripts
d = property or dimension
of source and drain
f = property or dimension
o f fin
g = property or dimension
o f gate
m = property or dimension
of metal interconnect
o = . property or dimension
o f silicon dioxide layer
AUGUST 1995, Vol. 117 / 575
Downloaded 21 Nov 2008 to 171.64.49.29. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
Table 1 Common dimensions of SOl FET devices and those of the test
structures used here
Dimensions
In FET
(~m)
In Test
Structure (ttm)
implanted-SiO2 thickness, do
additional-SiO2 thickness, de
interconnect thickness, dm
device thickness, dd
0.4
0.60
0.5
0.08
0.293 - 0.503
0.60
I
0.041 - 0.177
8ate thickness, dg
gate-channel separation, dgo
channel-interconnect separation, Ld
~ate half-width, Lg
device width, wa
channel-gate contact separation, We
0.30
0.0055
0.5
0.25
0.8
0.29
0.0055
0.8 - 3.8
0.16
10
Table 2 Thermal conductivities of SOl circuit materials. These values are
discussed in Section 2.2. The channel thermal conductivity was not used
by Goodson and Flik (1992), but is needed in Section 3.2 for the analysis
of the experimental data.
Thermal
conductivity at
To = 303 K
(Wm-I K-I)
Region or
Component
Material
substrate
single-crystal silicon,
3 x 1015 boron atoms cm -3
ks= 148"
channel
single-crystal sificon
6 x 1017 boron atoms cm-3
ke = 148*
source and
drain
single-crystal silicon,
1 x 1020 arsenic atoms cm-a
kd = 63*
gate
polysificon,
1 x 1020 arsenic atoms cm-3
kg = 30"*
aluminum,
km = 239*
interconnect
1 mass-percent silicon
only on the ratio of the width of the fin and the thickness of the
thermally resistive layer below the fin. The values of • used here
range between 1.016 and 1.028 for conduction from the source
and drain, between 1.051 and 1.064 for conduction from the interconnects, and between 2.349 and 2.925 for conduction from
the gate. The largest Biot numbers for the test structures in the
present manuscript, d:h/.Ik/, where k: and d/are the fin conductivity and thickness, are 0.028 for the source and drain fins, 0.029
for the gate fin, and 0.00782 for the interconnect fins. The fin
thermal healing length is 1/m = (kAt./hwr) =/2, where k and At
are the thermal conductivity and cross-sectional area of the fin.
The distance from a heating source over which the fin temperature recovers to the substrate temperature is of the order of the
healing length. The healing lengths are 1/m,,, = (k,,,d,,Ih,,,)J/2
7 # m for the interconnects, 1/md = (k, dd/h,)~/z ~ 1 #m for the
source and drain, and 1/m~ = (k~d~/h.~)~/2 ~ 1.5/zm for the gate.
The length in the X direction of the contact between the interconnect and the drain is approximately 3 /zm. Because this
length is substantially smaller than the healing length of the
interconnect fins, Goodson and Flik (1992) neglected the different heat transfer coefficient between the interconnect and the
substrate over this length. A single fin is used for the interconnect and the contact region with the heat transfer coefficient
that prevails when the interconnect is above the thermal and
C V D silicon-dioxide layers. This slightly overestimates the
thermal resistance between the channel and substrate, but substantially simplifies the analysis.
The width of the interconnect, source, and drain fins in the Y
direction, 10 #m, is larger than the healing length in these fins.
This makes possible significant temperature variation in this
direction, which is not considered by the present fin analysis.
Temperature variation due to conduction from the sides of these
fins is small because the thermal resistance for conduction by a
unit area through one healing length in the fin, approximately
(dl,,d~) ~/2/(k,,k:)i/2, is much smaller for each fin than the thermal resistance for conduction from a unit area of the fin side to
the substrate, approximately dr,~k,,. But temperature variation
in the Y direction due to conduction through the gate fin is
significant. The error is estimated using an analytical solution
to the three-dimensional thermal-conduction equation in the implanted silicon-dioxide layer and the channel, source, and drain.
The difference between the conductivity of the channel and that
of the source and drain is neglected and the source and drain
fins are assumed to be very long in the X direction. Heat generation in the channel is modeled using a uniform heat-flux
boundary condition at the top surface of the channel. Heat loss
to the gate is modeled using a heat flux condition from the top
of the channel near the edge, wd - dd --< ] Y] --< wa. The average
channel temperature rise is slightly larger than those calculated
by neglecting temperature variation in the Y direction in the
576 / Vol. 117, AUGUST 1995
SOI implanted
layer
silicon dioxide,
implanted
ko = 1.40"* *
other
insulating layers
silicon dioxide,
thermally ~rown and CVD
ko = 1.40"**
*Touloukian et al. (1970)
**Tai et al. (1988)
***Sugawara (1969)
channel, which is the approximation of the simple multifin analysis used in this manuscript. The relative error in the predictions
given here is less than 3 percent for a channel width wa = 0.3
# m and less than 7 percent for w,~ = l0 #m.
The devices were assumed to be in an infinite linear array,
each connected by an interconnect of length 2L,,,, and each dissipating the same power P. This idealization resulted in an estimate of the worst-case temperature distribution in a real circuit
for a given value of the device separation, 2L,,. It yielded the
two planes of symmetry shown, which were adiabatic boundaries. The temperature and location in the interconnect are given
by Tm and the parameter x,,,, in the drain by Ta and Xd, and in the
gate by T~ and xe. The parameters x,,,, Xd, and xe are not related
to the coordinates X, Y, and Z. The channel temperature is T,.
The gate is separated from the channel by a silicon-dioxide layer
of thickness d~,, = 5.5 nm, whose thermal resistance is negligible,
yielding T~.(x~. = 0) = T,.. The fin equations were solved by
requiring temperature continuity and energy conservation at the
fin interfaces, yielding
(1)
T,, - To = Z, cosh [m,,(L,, - x,,)]
Td - To = Z2 exp[mdx, t] + Z3 e x p [ - m . x . ]
W
d
,d
DRAIN"
Fig. 2
1992)
,
e~"~,I
2Lgl
I ~Ld~
L ~
A~
(2)
GATE
| " Idg
--Am
i
PLANESOF
SYMMETRY
I
I
Geometry of the thermal model of a SOl FET (Goodson and Flik,
Transactions of the ASME
Downloaded 21 Nov 2008 to 171.64.49.29. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
T~ - Ti, = Z4 exp[-m~x~l
(3)
The variables Z,, Z2, Z3, and Z4 were determined by solving a
set of four simultaneous algebraic equations, given in matrix
form by Goodson and Flik (1992). The channel temperature is
T, = T~(x~ = O) = Z4 + To. The maximum temperature in the
interconnect, at the interconnect-device contact, is T,........, =
T,,,(x,, = O) = Z2 + Z3 + To. The channel-to-substrate thermal
resistance is R, = Z J P .
The isothermal-channel approximation does not account for
the strong spatial dependence of the rate of heat generation in the
channel (e.g., Ostermeier et al., 1992; Fushinobu and Majumdar,
1993). The greatest error occurs when the transistor is saturated,
which results in heat generation predominantly near the channeldrain interface. An upper bound for the error is estimated using
the fin method of Goodson and Flik (1992) and the assumption
that all of the heat generation occurs within a region at the channel-drain interface of length in the X direction two orders of
magnitude less than the channel length, L J 5 0 . Conduction to the
gate and the difference between the channel thermal conductivity
and that of the source and drain are neglected. This approximate
analysis of the error yields a highly nonuniform channel temperature rise. But the a v e r a g e channel-temperature rise differs by
less than 0.5 percent from that predicted by assuming an isothermal channel. The analysis of Goodson and Flik (1992) is
therefore appropriate for predicting the average channel temperature rise, which is measured by the experimental technique developed here. In contrast, this analysis of the error shows that the
isothermal-channel approximation results in a significant underprediction of the temperature rise of the drain-interconnectcontact. For L,~ = 0.5 and 1.5 #m, the relative errors are 13 and 11
percent, respectively.
The analysis neglects thermal boundary resistances. Goodson
et al. (1994) measured the effective thermal conductivities for
conduction normal to the implanted silicon-dioxide layers in the
present study bounded below by silicon and above by aluminum.
The data agree closely with the conductivity of bulk silicon dioxide, making a significant thermal boundary resistance between
the implanted layer and the silicon dioxide unlikely. Data are
needed for the boundary resistance between the buried silicon
dioxide and the silicon device, as well as the resistances at the
boundaries of the thermally grown silicon dioxide. Goodson et
al. (1993) measured the effective conductivities for conduction
normal to chemical-vapor-deposited layers fabricated using the
same method as those in the present study. They obtained upper
bounds for the thermal boundary resistance between the aluminum and the silicon dioxide that would not significantly change
the values of hr for the interconnects in the present work. The
thermal resistances that impede conduction from the channel to
the gate and from the source and drain to the interconnects are
unknown. If significant, these resistances will cause the analysis
here to underpredict the channel-to-substrate thermal resistance.
2.2
The source and drain are single-crystal silicon doped with approximately 1 × 102o arsenic atoms cm -~. The most appropriate
existing data are for bulk single-crystal silicon doped with 1.7 ×
10 z~ phosphorus atoms cm 3 (Touloukian et al., 1970). There
are no thermal conductivity data available for the gates in the
present research, which were polysilicon heavily doped with arsenic atoms. Tai et al. (1988) and V61klein and Baltes (1992)
measured thermal conductivities near 30 W m-~ K ~ in polysilicon layers heavily doped with phosphorus. The data are consistent with the thermal diffusivity data of Mastrangelo and Muller
(1988), also for phosphorus-doped polysilicon layers. The conductivity k~ = 30 W m ~ K - ' is used here. The conductivity
above 300 K of silicon with less than 10 ~x dopant-atoms cm 3
differs little from that of intrinsic silicon (Touloukian et al.,
1970), which is used for k, and k~ here. The thermal conductivity
of aluminum layers containing 1 mass percent of silicon has not
been measured directly. The thermal conductivity calculated using the Wiedemann-Franz law (Kittel, 1986) and the electrical
resistivity measured here of these layers is within 4 percent of
the thermal conductivity recommended for bulk aluminum (Touloukian et al., 1970), so thebulk value is used here.
The substrate temperature during the measurements was T~ =
303 K, and the largest channel temperature was T, = 403 K. The
thermal model in Section 2.1 neglects the temperature dependence of the thermal conductivities of the SOl FET materials.
This is a good approximation between 303 and 433 K for silicon
dioxide and aluminum, whose bulk thermal conductivities vary
by less than 13 and 2 percent in this range, respectively. But the
thermal conductivity of the heavily doped silicon source and
drain varies more significantly in this temperature range. To help
overcome this difficulty, the temperature (T, + T,)/2, averaged
for all of the data, is used when interpolating k,t.
Dimensions.
Figure 3 compares the experimental test structure with a FET device. The measured dimensions of the test
structures are given in Table l. The interconnect lengths are very
long compared to 1/m,,,, so that L,,, = ~ is used. The gate of the
FET in the test structure extends out from both sides of the channel to interconnect contacts in the Y direction. This results in a
plane of symmetry normal to the Y axis, i.e., an X - Z plane, at Y
= 0. The temperature field in each half of the test structure
is predicted by the thermal model. The thermal analysis is applied using w,i = (w,/)t~t......... /2, W., = (W,,,),cststructurc]2. and P =
(P) ..... ,....... /2.
s
ERCOI~LCI'"")"
Application to the Experimental Test Structure
T h e r m a l Conductivities. Table 2 gives the thermal conductivities used in the model. Goodson et al. (1994) measured the
thermal conductivity of implanted-silicon-dioxidelayers in SOI
wafers near room temperature. The data agreed within the experimental error with the value recommended for bulk amorphous silicon dioxide (Sugawara, 1969), so the bulk value is
used here. The bulk value is also used for thermally grown silicon
dioxide, supported by the data of Goodson et al. (1993). The
thermal conductivities reported for chemical-vapor-deposited
(CVD) silicon-dioxide layers do not agree, but are in general
less than the bulk value (Schafft et al., 1989; Brotzen et al., 1992;
Goodson et al., 1993). Due to the lack of a consensus among the
data for CVD silicon-dioxide layers, the bulk value is used in
this case as well. The resulting error is very small because the
thermal resistance of the implanted silicon-dioxide layer is much
more important than that of the CVD layers.
Journal of Heat Transfer
+
VG
VDs
a) FET DEVICE.
b) TEST STRUCTURE.
Fig. 3 Top views of: (a) FET device; (b) test structure
AUGUST 1995, Vol. 117 / 577
Downloaded 21 Nov 2008 to 171.64.49.29. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
Parameter Uncertainties. An uncertainty in the predictions
of the thermal analysis results from the use of parameters, e.g.,
thermal conductivities and dimensions, which may differ from
those in the test structure. This uncertainty is estimated using the
sum-of-squares technique (e.g., Holman, 1984) and the analytical model. The largest expected error is ±10 percent. The uncertainties in k,~ and da are the largest contributors.
Heat Transfer to Ambient Air. The test structure is exposed
to ambient air, but heat transfer to the air is neglected. This is
justified by the small value of the channel-to-air thermal conductance compared to the channel-to-substrate thermal conductance,
which is predicted in this work to be 11R, ~ 0.5 - 2 x l 0 -4
W K ~. The channel-to-air thermal conductance is of the order
of that from an isothermal disk of radius a on the boundary of a
semi-infinite medium of conductivity k, G = 4ak (Carslaw and
Jaeger, 1959). Using a = wd/2 and the room-temperature conductivity of air yields G = 5.2 x 10 -7 W K -I . An order-ofmagnitude analysis of the momentum equation estimates the air
velocity near the device due to buoyancy forces (Rohsenow and
Choi, 1961 ), v = pga213(T, - To)~#, where g is the acceleration
due to gravity, and p is the density, # is the viscosity, and/5
1~To is the approximate coefficient of thermal expansion of the
air. Using room-temperature properties and T,. - To = 130 K
yields v = 6.6 #m s ~. The thermal conductance contributed by
the air motion is of the order of G = vTra2pc~, = 6.2 X 10 -~3
W K ~, where c~, is the specific heat per unit mass at constant
pressure of air. The thermal conductance due to radiation is of
the order of G = 4e~rTi]A, where e is the emissivity of the surface,
is the Stefan-Boltzmann constant, and A is the area of the
emitting surface. Using e = 1 and A = 7r a 2 yields G =
5.0 x 10 m W K -t at room temperature.
3
Channel-Temperature Measurement Technique
This section describes the technique for measuring the channel
temperature of SOl FETs. Section 3.1 describes the apparatus
and the general procedure, and Section 3.2 calculates the channel
temperature from the measured gate resistance. Section 3.3 determines the experimental uncertainty.
3.1 Apparatus and Procedure. Figure 3 shows the experimental structure. The electrical resistance of the gate depends
strongly on temperature. It serves as an electrical-resistance
thermometer. The calibration measures the gate electrical resistance, R(;, as a function of temperature when there is no drain
current, i.e., when the gate is isothermal, yielding
[Rc,(T)]~,,ub~,,~..... The substrate temperature is controlled using
a Temptronic Model TP38B chuck, a copper disk with a diameter of 88.9 mm and a thickness of 19.1 mm, to which the
wafer is secured by suction. A thermocouple with one junction
soldered to the chuck surface measures the chuck temperature.
The chuck is maintained at the temperature To and the gate
resistance is measured for varying values of the drain-source
voltage drop, Vt)s, and the gate-source voltage drop, V(;s, i.e.,
for several different device powers, P = loVos.
The average gate temperature is defined as that of the gate
segment whose resistance is measured, i.e., the segment between
the voltage contacts. The average channel temperature is T,.,
which is shown in Section 3.2 to be very well approximated by
the average temperature of the gate segment over the channel.
The average gate temperature considers the gate segments not
over the channel heater, and is less than T,. The FET gate-temperature variation is more important in a SOI wafer, where most
of the temperature drop occurs within a few micrometers of the
channel due to the implanted layer, than in a normal substrate,
where the temperature-drop length scale is the thickness of the
substrate, i.e., a few hundred micrometers. This temperature variation must be considered when calculating T, from R~;.
3.2 Temperature Distribution in the Gate. The gate temperature variation in the Y direction is
578 / Vol. 117, AUGUST 1995
T(;(Y) = T, + AT(;F(;(Y)
(4)
where AT(; is the average gate-temperature rise from T, and
F~;(Y) is a shape function of average value unity that is defined
for ] Y] < w,, + wj/2. For each measured R(;, ATe; is determined
iteratively using
Rc;
=
f•,
.,.+,,,,12
,,.,. ,,.,,/2
dY
[R~;(To + AT~;F~;(Y))]~.,,lib,.,,,i,,,, 2w,. + wd
(5)
The thermal resistance of the silicon-dioxide layer between the
channel and gate, dv,/(2L~w,lk,,) = 1 . 2 x 103 K W-~, is small
compared to the thermal resistance for conduction along the gate
to the contact in the Y direction, w,,/(2L~,d.~k~ ) = 7.31 x 105 K
W -~. This means that the channel- and gate-temperature distributions are almost identical for I Y [ < wfl2. The average channel
temperature is
T, = To +
f ,,,/2
dY
AT~F(;(Y)-
wdl2
(6)
Wd
The channel-to-substrate thermal resistance is R, = (T, - To)/P
= ( T , - To)l(l,)V,,s).
Two shape functions are now developed, from which Eqs. (5)
and (6) yield upper and lower bounds for T, for a given Ru. Each
shape function must be even, due to the test-structure symmetry
about Y = 0, and continuous. Because of the large width in the
X direction of the gate-interconnect contacts, 4 #m, compared to
the gate width, 2L.~ = 0.32 #m, the contacts are very nearly isothermal at the substrate temperature, T0. This yields the boundary
conditions FG(Y) = 0 at Y = ±(w,. + w,J2). For a given Re,, T,
calculated using Eqs. (5) and (6) increases with the difference
between unity and the average of Fu(Y) over - w , J 2 < Y < wjI
2, i.e., with the assumed difference between the average gate and
channel temperatures. Shape LB assumes a linear temperature
profile in the gate segments not over the channel, which neglects
conduction down through the buried silicon-dioxide layer, and
an isothermal channel. Both assumptions underestimate the difference between the average channel and average gate temperatures, yielding a lower bound for T,.. Shape LB is
+
Fa(Y) - Wd
w,, + w,.
2w------s,
FG(Y) =
IYI
<--
Wd
2
(7)
[ wa + 2w,,] [(w,, + wa/2) - [Y[]
w,i + w,, j
w,.
'
Wd
Wd
--<2 IYl<w,+ T (8)
An overestimate of the difference between the average gate
and channel temperatures requires an overestimate of the temperature drop between the center and the edge of the channel in
the Y direction. This is calculated by isolating the channel and
gate from the source and drain, in which thermal conduction
reduces the channel-temperature variation in the Y direction. The
gate and channel are grouped together as a composite fin, which
meets a fin of different internal properties and heat transfer coefficient at Y = w,J2. Solving the heat equation in the two fins
yields shape UB, which has a larger average in the channel region
than shape LB. Values of T,. - To calculated using shapes UB
and LB differ by less than 8 percent, and the simpler shape LB
is used here, Eqs. (7) and (8). This function does not describe
the temperature distribution in the channel. Rather, it is a shape
function which, when used in Eqs. (5) and (6), yields a value
for T, close to the average channel temperature. The difference
between the upper and lower bounds is used in the experimentaluncertainty analysis.
3.3 Experimental Uncertainty. The uncertainty in
R, = ( T, - T o ) / P has three significant, independent components: (a) There is a relative uncertainty of 4 percent in
Transactions of the ASME
Downloaded 21 Nov 2008 to 171.64.49.29. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
440
~
420
•
i
•
i
,
i
.
i
.
i
L d = 1.5 Iam
.
r
20
.
'~'
4~
~ d d = 41 n m
'
'
\
~
J
'
. . . .
'
. . . .
d: ='35;
:rn'
L d = 1.5 pm
Wd= 1 0 p m
• =
~
~A d d = 78 n m
soinEV,CES
,O •
d o = 0.359 lam ,,"
~
,ll
Z
=
.O d d = 177 nm
Z
340
•
,O" "
5.0
BULK
DEVICE
- []
PREDICTED
MEASURED
320
ii
0.0
3oo
i
" I
2
i
i
i
4
I
6
i
i
8
i
I
10
i
I
12
511
101)
151)
200
i
14
DEVICE THICKNESS,
d a (nm)
D E V I C E P O W E R , P (roW)
Fig. 4 Channel-temperature data for SOl FETs with varying device thicknesses, compared with data for a bulk (non-SOl) FET
T, - To due to the error in the substrate-temperature change
measured by the chuck thermocouple (Goodson et al.,
1994). (b) A relative uncertainty of 6.6 percent in T, - T0
is due to the measurement of Re,. (c) A relative uncertainty
of 8 percent in T,. - To is due to the approximate shape
function for the temperature profile in the gate, as shown in
Section 3.2. The total relative uncertainty in R, is ± 11 percent, determined using the sum-of-squares technique (Holman, 1984).
4
Results and Discussion
Section 4.1 compares data for the channel-to-substrate thermal
resistance of SOI FETs with predictions of the thermal analysis
of Section 2. The thermal analysis is used in Section 4.2 to estimate the influence of the implanted silicon-dioxide layer on the
reliability of highly integrated SOI circuits.
4.1 Channel-to-Substrate Thermal Resistance. Channeltemperature measurements are performed on SOI test structures
with varying values of L,i, d,~, and d,,. Test structures fabricated
from bulk (non-SOI) wafers are measured for comparison. The
device voltages satisfy 0 V < V~s < 3 V and V~s = 2 V and 2.5
V, which are typical operating conditions. The device powers
vary between 3 and 14 mW, and the values of T, - T0 vary
between 5 and 130 K. The power dissipated in the gate electricalresistance thermometer is at least two orders of magnitude
smaller than the device power.
The channel-to-substrate thermal resistance, R,, varies by less
than the experimental uncertainty for varying powers from a single device, as shown in Fig. 4 for three SOI devices and one bulk
device, i.e., a device fabricated in a conventional wafer, which
lacks the implanted silicon-dioxide layer in SOI wafers. The data
for each device for varying powers fall near a line originating at
P = 0 and T0 = 303 K, whose slope is R,.. Values of R,. for the
SOI devices are as much as 10 times larger than R, for the bulk
device, due to the thermal resistance of the implanted silicondioxide layer. The value of R, decreases with increasing SOI
device thickness, da. In Figs. 5 and 6, each data point is the
average of the values of R,. measured in a single test structure.
Uncertainty bars are only given for selected data to avoid cluttering of the figures. Each data point is the average of the values
of R, measured in a single test structure. Uncertainty bars are
only given for selected data to avoid cluttering of the figures.
Figure 5 shows that the sensitivity of R,. to the device thickness
is predicted by the thermal model of Section 2. Increasing d,~
reduces the channel temperature for a given power. The parameters ka and da are not independent, but always appear as a product in the solution for the temperature distribution (Goodson and
Flik, 1992). Thus, the channel temperature is also sensitive to
kd, which depends on the doping level in the source and drain.
Journal of Heat Transfer
Fig, 5 The channel-to-substrate thermal resistance, R~, as a function of
the device thickness
The agreement is excellent considering the uncertainties of the
thermal conductivities and dimensions used in the analysis• Section 2.2 estimated that the potential relative error in the predictions due to these uncertainties is ± 10 percent• The differences
in the experimental data at device thicknesses near 45 #m in Fig.
5 do not indicate a poor repeatability of the experimental method.
Each data point is for a different experimental structure. Because
R,. depends sensitively on the structure dimensions, the uncertainty in the measurement of structure dimensions can yield data
on a single graph that are apparently inconsistent when only the
uncertainty in the measurement of R, is considered•
Figure 6 shows the dependence of R, on the implanted-silicondioxide layer thickness. The data support the predictions of the
model, and indicate that R, is as sensitive to d,~ as it is to d,,. This
is in contrast to the predictions of McDaid et al. (1989), whose
model assumed that R,. is independent of d,~. By modeling onedimensional conduction in the implanted-silicon-dioxide layer,
these researchers predicted that R, = d,,/(Ak,), where A is the
device area in the X - Y plane. This neglects the spreading of the
temperature profile into the source and drain fins with increasing
d,,, and is not consistent with the data. This can be remedied by
a simple scaling analysis• The area in the source and drain with
significant temperature rise is of the order of A = 2wd/re,t, where
the thermal healing length 11md is approximately (k,ld, td,,Ik,,)~/2.
Using R, = d,,/(Ak,) with this expression for A yields
l ( d,, l '/2
R,. ~ 2w,---~\ k,,~,td~, /
(9)
which is in qualitative agreement with the data in Fig. 6. This
shows that R,. is roughly proportional to d~/2, and that the sensitivity of R, to d,, and d,~ is similar, i.e., halving d,, and doubling
20 ~ ' ' ' l ' ' ' ' l ' ' ' ' l ' ' ' ' l
....
I'''
'1
' '''
CIIANNEL T F M P E R A T U R E :
PREDICTEI), d.~ = 54 nm
.....
<~
15
PREI)ICTEI), d] = 177 nm
MMiEAA~URED, d d = 5 4
nm
T
-
i
0.0
0
,,I,,,,l~h,,I
ILl
....
0,2
0.3
[,,,,I,,,,i,,,
0.4
0.5
l),6
0.7
IMPLANTED SILICON-DIOXIDE
THICKNESS, d (pro)
o
Fig. 6 The channel-to-substrate thermal resistance as a function of the
implanted silicon-dioxide layer thickness
A U G U S T 1995, Vol. 117 / 579
Downloaded 21 Nov 2008 to 171.64.49.29. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
d,~ have the same impact, if all other parameters are held constant.
Equation (9) is valid only when the thermal healing length in the
source and drain is smaller than the channel-interconnect separation, i.e., 1/m,~ < La. Otherwise, there is significant heat conduction into the interconnects, which are more effective fins than
the source and drain because of their large thickness and high
thermal conductivities.
4.2 Eleetromigration-Limited Reliability of FET-Intercon.
nect Contacts. The FET-interconnect contact temperature increases as L,~ is reduced, because this brings the contact nearer
to the channel heat source. This reduces the reliability of the
contact, whose electromigration-limited mean time to failure
(MTF) decreases with increasing temperature. This section estimates the reduction of the MTF of FET-interconnect contacts
due to the implanted layer in SOl circuits and demonstrates the
potential for overcoming this problem through transistor-level
thermal design. Electromigration is the motion of metal atoms in
the interconnect in the direction of electron flow due to electronlattice momentum transfer. This causes interconnect failure due
to void formation, particularly near FET-interconnect contacts,
where the flux of metal atoms diverges. The temperature dependence of the MTF limited by electromigration is (e.g., Black,
1967)
MTF = K,. exp k ~
(10)
where E~, is the activation energy for atomic diffusion,
k~ = 1.38 × 10 -23 J K J is the Boltzmann constant, T is the
temperature, and Ke is a function of the electrical current density,
the geometry, and the microstructure and purity of the metal.
Equation (10) agrees well with data for FET-interconnect contacts ifE,~ = 0.5 eV = 8 × 10 -2° J is used (Chern et al., 1986).
The FET device dimensions in Table 1 are used with an interconnect length between devices of 2L,, = 4 ,am. The FETs
experience steady-periodic heating with pulses of 1.61 W for one
tenth of each clock cycle, where 1.61 W is the steady-state power
of a SOI device with these dimensions (Woerlee et al., 1989).
The analysis in Section 2 provides a good estimate of the nearly
steady-state FET-interconnect contact temperature if the timeaveraged power is used, P~vg = 0.161 W. The resulting contact
temperature rise is less than 8 K, and depends strongly on d,, and
Let. Figure 7 uses Eq. (10) to show the ratio of the MTF for FET interconnect contacts in a SOI circuit to that for contacts in a
bulk circuit. These predictions assume that the substrate temperature in the SOl case is equal to that in the bulk case. The difference between the two contact temperatures is due to the thermal resistance of the implanted silicon-dioxide layer. The MTF
increases as La is increased, because the interconnect moves away
from the channel heater. It may be possible to improve circuit
reliability by increasing L,~, but this must be weighed against the
need for compact devices. Reducing d,, also increases the MTF,
because this reduces the contact temperature.
Figure 7 provides the type of information needed to make decisions effectively during the design of SOl circuits,but the predictions are very approximate. Because Eq. (10) has only been
experimentally verified using accelerated testing, i.e., the use of
electrical current densities and temperatures which are higher
than those found in an operating circuit, Fig. 7 can at best show
the expected trends. In a real device, the rate of heat generation
will be greatest near the drain contact, causing this contact to
have the lowest MTF. But the isothermal-channel approximation
of Goodson and Flik (1992) yields identical temperatures in the
source and drain contacts. As discussed in Section 2.1, this underpredicts the temperature rise of the drain contact by as much
as 13 percent. The resulting error in the normalized MTF in Fig.
7 is less than 10 percent.
5
Conclusions
The analysis of Goodson and Flik (1992) tends to underpredict
the data presented here. This is due in part to the assumption of
580 / Vol. 117, AUGUST 1995
0.9
.
I"ET IN INTI';GRATI';I)CIRCUIT,
I)IMENSIIINS 1N TABLE 1 ANI)
2L m = 4 /urn, I'0 = 300 K
/. ~
Pale = 0.16 mW
/ /
Et = 0.5 eV
/ / /
~ ~
~
t'7~
11.7
/
/
0,6
~,
~
-'~
1,;n
~s" snM.~ON-IHOXU)E
THICKNESS
-----do=0.1
~m
.''"
0,5
t
I ,''1
0,5
I
d o = 1.0 /urn
I
t I
t
CHANNEl:INTERCONNECT
L d (~nl)
.....
d o =10.4 prn
SEPARATION,
Fig. 7 Predicted dependence of the median time to failure (MTF) of FETinterconnect contacts on the channel-interconnect separation
an isothermal channel, which results in about a 7 percent underprediction of the channel-to-substrate thermal resistance, as discussed in Section 2.1. The underprediction may be due in part to
thermal boundary resistances in the structure, which, as discussed
in Section 2.1, are not considered by the analysis. Another important possibility is that the thermal conductivities of the interconnects and source and drain, which have not been measured,
may be less than the bulk values used here. Better agreement will
require more research on thermal conductivities and boundary
resistances in circuits.
The mean time to failure of a circuit and the channel mobility
are important to the design of circuits. They are affected by the
channel and interconnect temperatures, which are shown here to
depend strongly on design parameters, e.g., k,/, da, d,,, and L,~.
Some of these parameters also affect the electrical performance
of the device directly. In order to achieve circuits of optimal
performance and reliability, design for electrical performance
should be accompanied by device thermal design, i.e., the choice
of dimensions, materials, and processing techniques that enhance
heat conduction within a few micrometers of the device. This
work provides a basis for the thermal design of SOI FETs.
This work shows that the steady-state channel-temperaturerise
in SOl FETs due to Joule heating is significant. Electrical-property measurements performed on devices in the steady state are
affected by this temperature rise due to the strong dependet~ce of
the channel mobility on temperature. The steady-state data may
not be applicable to devices in an integrated circuit, where the
channel-heating is time dependent. More work is needed to determine time-dependent temperature fields in integrated SO1 circuits.
Acknowledgments
Professors Carl V. Thompson and Borivoje B. Mikic of M.I.T.
provided valuable comments. K.E.G. and L.T.S. were supported
by academic fellowships from the Office of Naval Research and
AT&T Bell Labs, respectively. The SOl wafers used in this study
were provided by IBIS Corp., Danvers, Massachusetts. D.A.A.
and L.T.S. acknowledge the support of the Semiconductor Research Corporation (SRC) under contract No. 92-SP-309.
References
B l a c k , J. R., 1967, " ' M a s s T r a n s p o r t o f A l u m i n u r n by M o m e n t u m E x c h a n g e W i t h
C o n d u c t i n g E l e c t r o n s , " Proceedings of the IEEE Reliability Physics Symposium,
pp. 1 4 8 - 1 5 9 .
Brotzen, F. R., Loos, P. J., and Brady, D. P., 1992, " T h e r m a l C o n d u c t i v i t y o f
T h i n SiO~ F i l m s , " Thin Solid Fihns, Vol. 2(]7, pp. 1 9 - 7 2 0 1 .
B r u g g e r , H., 199 I, " R a m a n S p e c t r o s c o p y ['or C h a r a c t e r i z a t i o n o f L a s e r e d S e m i c o n d u c t o r M a t e r i a l s and D e v i c e s , " in: Light Scattering in Semiconductor Structures
and Superlattices, K. J. L o c k w o o d and J. F. Y o u n g . eds., P l e n u m , N e w Y o r k , pp,
259-274.
Transactions of the ASME
Downloaded 21 Nov 2008 to 171.64.49.29. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
Bunyan, R, J. T., Uren, M. J., Alde,'mau, J. C., and Eccleslon, W., 1992. "Use
of Noise Thermometry to Study the Effects of Self-Heating in Submicron]eter SOl
MOSFET's," IEEE Electrml Device Lettelw, Vol, 13, pp. 279-281.
Carslaw, H, S., and Jaeger, J. C., 1959, Comhlctiml qf Heat in SoIMs, Oxford
University Press, New York, pp. 214-216.
Chern, J, G. J., Oldham, W. G., and Cheung, N., 1986, "Electromigration in All
Si Contacts--Induced Open-Circuit Failutv," IEEE Transactions mt Elec/twl Devices. Vol. ED-33, pp. 1256 1262,
Colinge, J. P., 199 I, Silicon-on-h*suhttor Teehm~logy: Materials to VLSI, Kluwer
Academic Publishers, Boston, pp. I-5.
Fushinobu, K., and Majumdar, A., 1993, "Heat Generation and Transport in Submicron Semiconductor Devices," in: Heat 7)'an,~ybr on the Microscah', ASME
HTD-Vol. 253, F. M. Gerner and K. S. Udell, eds., pp. 21-28.
Goodson, K. E., and Flik, M. l., 1992, "Effect of Microscate Heat Conduction
on the Packing Limit of Silicon-on-Insulator Electronic Devices," IEEE Trans. on
Components. Hybrids, and Mant~u'turing Tech~tology, Vol. 15, pp. 715-722.
Goodson, K. E., Flik, M. l., Su, L. T., and Antoniadis, D. A., 1993, "AnnealingTemperature Dependence of the Thermal Conductivity of CVD Silicon-Dioxide
Layers," IEEE Electtvn Device Lette~w, Vol. 14, pp. 490-492.
Goodson, K. E., Flik, M l., Su, L. T., and Antoniadis, D. A., 1994, "Prediction
and Measurement of the Thermal Conductivity of Amorphous Dielectric Layers,"
ASME JotJl~NM, oJ: HI,AT Tt,tANS{-r~{~,Vol. 116, pp. 317 323.
Hohnan, J. P., 1984, Experimentol Methods.[~r Engineetw, McGraw-Hill, New
York, pp. 50-57.
Kittel, C., 1986, Introduction to Solid State Physics, Wiley, New York.
Lai, J., Carrejo, J. P., and Mt~iumdar, A., 1993, "Thermal hnaging and Analysis
at Sub-micrometer Scales using the Atomic Force Microscope," in: Heal Tran,v[~,r
on the Microscale. ASME HTD-Vol. 253, F. M. Gerner and K. S. Udell, eds., pp.
1 3 - 20.
Lifka, H., and Woerlee, P. H., 1990, "Thin Simox SOl Material for Half-Micron
CMOS," Proceedings (~["the European Solid Sutte Device Reseatz'h Cot!/~'rence,
pp. 453-456.
Majumdar, A., Carrc~}o,P,, and Lai, J., 1993, "Thermal hnaging Using the Atomic
Force Microscope," Appl. Phys. Lett., Vol. 62, pp. 250t-2503.
Marcus, R. B,, and Sheng, T. T., 1983, Transmission Electron Mie~vseopy (t['
Silicon VLSI Circuits and Structures, Wiley, New York, pp. 80-82.
Journal of Heat Transfer
Mastraugelo, C. H., and Muller, R. S., 1988, -Thermal Diffusivity of Heavily
Doped Low Pressure Chemical Vapor Deposited Polycrystallinc Silicon Filnas,"
Sensors emd Materia&, Vot. 3, pp. 133-142.
Mautry, P. G., and Trager, J., I t~)90,
¢
• and Temper tu "c Mc st "c ~ac it
SelI-Hea mg
in Sub-,um-MOSFETs," Proceedings el/the IEEE Interllaliom/I Cmt[ercnce ott Microelectrot/ic Test Strttcttoes, Vol. 3, pp. 221 226.
McDaid, L. J., Hall, S., Mellor, P, H., Eccleston, W., and Alderman, J. C., 1989,
,Physical Origin of the Negative I)iffercntial Resistance in SOl Transistors," Electroni~w Lett., Vot. 25, pp. 827-828.
Ostermeier, R., Brunner, K., Abstreiter, G., and Weber, W., 1992, "Temperature
Distribution in Si-MOSFET's Studied by Micro Raman Spectroscopy," IEEE
Tra/ts. E/ectro/t Del'ices. Vol. 39, pp. 858-863.
Rohsenow, W. M., and Choi, H. Y., 196 I, Heat, Mass, ~md Mome/tlutn Tran.sfi'/,
Prentice-Hall, Englewood Cliffs, NJ, p. 155.
Rowe, D, M., and Bhandari, C. M., 1986, "Preparation and Thermal Conductivity
of Doped Semiconductors,'" Progress in Ct3,xtal Growth and Characterization, Vol.
13, pp. 233-289.
Schafft, H. A., Suehle, J. S., and Mirel, P. G. A., 1989, "Thermal Conductivily
Measurements of Thin-Fihn Silicon Dioxide," Proceedings ~/the IEEE Internatiomd Cm![i~rem'e on Mictwelectronic Test Structures, Vol. 2, pp. 121 - 124.
Sugawara, A., 1969, "Precise Determination of Thermal Conductivity of High
Purity Fused Quartz fi'om 0 ° to 650°C, ' ' Physica, Vol. 41, pp. 515-520.
Tai, Y. C., Mastrangelo, C. H., and Muller, R. S., 1988, "Thermal Conductivity
o f Heavily-Doped Low-Pressure Chenfical Vapor Deposited Polycrystalline Silicon
Films," J. Appl. Phys., Vol. 63, pp. 1442-1447.
Touloukian, Y. S., Powell, R. W., Ho, C. Y., and Klemens, P. G.. 1971), "Thermal
Conductivity: Metallic Elements and Alloys," in: Thermol)hysieal Properties et]'
Matter; Vol. I, IFl/Plenurn, New York, pp. 9, 326, 330, and 335.
V61klein, F., and Baltes, X. X., 1992, " A Microstructure for Mcasurement of
Thermal Conductivity of Polysilicon Thin Films," .I. Mieroelec/romeelumical ,~vsterns, Vol. I, pp. 193-196,
Woerlee, P. H., van Ommen, A. H., Lifka, H., Juffermaus, C. A. It., Pl@t, L.,
and Klaassen, F. M., 1989, "Half-Micron CMOS on Ultra-Thin Silicon on Insulator'," Proceedings ~[' the IEEE hlternational Eleettvn Devices Meeting, pp. 821 824.
AUGUST 1995, Vol. 117 / 581
Downloaded 21 Nov 2008 to 171.64.49.29. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm